Direct Neural Network Calculation of Residual Stresses from Wrapped Speckle Interferometry Phase Patterns

Keywords

neural networks; speckle interferometry; phase unwrapping; residual stresses; automation; noise-aware training.

Abstract

The speckle interferometry hole drilling method for residual stress determination demands expert manual phase analysis. The developed 1D-CNN, trained on synthetic data, automates stress calculation directly from noisy wrapped phase data, bypassing manual steps like unwrapping and filtering. Validation on independent test data shows strong agreement. This approach eliminates operator variability and ensures reproducible stress determination.

References

1. Lobanov L.M., Pivtorak V.A., Savitskii V.V., Tkachuk G.I. Using electronic speckle interferometry for the accurate determination of the residual stresses in welded joints and structural members // Weld. Int. – 2010. – 24(6). – 439–443. https://doi.org/10.1080/09507110903464804
2. Shutkevych O. Determination of residual stresses using high-density current impulses and laser interferometry // PhD thesis, E. O. Paton Electric Welding Institute, National Academy of Sciences of Ukraine, Kyiv. –